Stop writing prompts. Start graph engineering.

Graph engineering is the practice of expressing what you want from an AI as a graph of nodes and connections instead of a block of prose. Each node holds one idea with its own context — a definition, examples, files, images, links. The edges carry the relationships: what follows what, what depends on what, what feeds back into what. The graph is then compiled into a specification the model can read without guessing. Read the full guide →

Loop engineering is the practice of designing the system that prompts an AI agent instead of prompting it yourself, turn by turn: a loop that gathers context, acts, verifies the result and repeats until the goal is met. The term took hold among people building with coding agents in mid-2026. The loop still needs an unambiguous definition of done — which is exactly what a graph provides. Read the full guide →

You open a node and explain it out loud. nodalo transcribes the recording, priming the transcription with your board's title, the node's name and every other node title and connection label on the board — so your own vocabulary is transcribed correctly. The transcript then goes to a language model, which rewrites it into a precise definition: every requirement kept, filler and self-corrections removed, written to instruction standard in the language you spoke, with nothing invented. The model is also given the node's incoming and outgoing connections, so the definition fits the node's position in the graph. The raw transcript is kept alongside it.

No — nodalo is where the loop is designed and specified, not a runtime that executes it. You draw the cycle, name the backward edge and write down the evaluation criteria and stop condition; the compiled specification hands that to whatever actually runs the work. Making the loop explicit is the point.

Prompt engineering optimises the wording of a single message. Graph engineering optimises the structure of the whole request: the parts, their order, their dependencies and the evidence attached to each one. Wording still matters, but ambiguity in a long prose prompt usually comes from missing structure, not from a badly chosen verb.

No. You need nodes and arrows. The useful ideas — order, dependency, branching, cycles — are visible on the canvas without any formal notation. nodalo handles the traversal and ordering when it compiles your graph into a specification.

nodalo compiles the graph into a Markdown specification: a reading guide that tells the model how identifiers and arrows work, an inventory, a Mermaid flowchart of the structure, one section per node with its own connections, definition, notes, links, embedded file contents and image identifiers, the annotations marked as author commentary, and the full relationship list. The order is topological — dependencies before dependents — not the order you drew things. Every element has a stable identifier: n1 for nodes, n3-b1 for the first image on node n3, n3.n1 for a node inside a sub-graph. You can copy that specification, download everything as a ZIP where each attachment is renamed to start with its identifier, or connect an assistant directly over the Model Context Protocol.

It covers the same reflex — think visually on an infinite canvas — but it is built for a different output. A whiteboard produces a picture for humans. nodalo produces a machine-readable specification, so every node has a contract, every attachment has a name an AI can reference, and the export follows the graph instead of your drawing order.

Yes. The export is plain Markdown plus a named attachment bundle, so it can be pasted into any assistant. nodalo also exposes the board over the Model Context Protocol for assistants that support custom connectors.

The canvas is free to use in your browser. Voice transcription and automatic node definitions run on your own API key, so you pay the model provider directly and nothing sits in between.